Machine learning has become a pioneering field of research in recent years. Since 2014, many big companies have invested in large data/machine learning departments complemented by serious investment in research. According to Arthur Samuel, machine learning gives computers the ability to learn without being explicitly programmed – Artificial Intelligence (AI) by any other name.
How do machines learn?
The learning process involves the creation of algorithms able to improve themselves, act and reason through data input. A stark example of machine learning and an example of the long way it has come since the ’80s is the ad policy in Google Search and the Facebook algorithms. Ads are no longer based on basic information typed in your profile or research but are chosen through compiled info based on indirect inputs such as likes, random small checks while researching and even interactions through emails. While one can debate the ethics implied in such processes as well as the future it implies for AI, we will now take a look at how this concept can be applied to mechanical design and its potential impact.
Machine learning and mechanical design
If by definition, machine learning is the insight elaborated through computers using input data, mechanical design engineering will be a potent field for this technique. Senior mechanical design engineers have built significant knowledge through several case studies and efficient working practices (plus a significant set of combined efficient learning, savant extrapolation and analysis mindsets). This creates the ability to approach new projects and be able to either harness their technical expertise with familiar elements that have already been unearthed or to use their experience to learn and adapt to new challenges. Rookies in the field learn and compile their efficiency and knowledge this way: as far as learning goes, mechanical design engineering is the one field where the a large chunk of learning is done on the job. How well its payoff will depend on one’s capacity to compile and process information.
It is therefore no wonder that many are adapting machine learning to perform such tasks and allow rookies to peek into detailed case studies depending on the models or requests they are facing. By applying algorithms on content-based databases and filtering the methodologies of design and highlighting features, machine learning can allow a mechanical design engineer to access the case studies best suited to their own project. This can provide invaluable advice about how to conduct the modeling of the part or the assembly.
Moreover, this also allows further extrapolated information based on the use of the results: once their use is documented, directly or indirectly, the algorithm is able to predict the most used techniques and reasons why they were preferred. The system would be able to offer a first assessment directly to the designer by assessing the requirements of the customer and giving a first recommendation. Needless to say, depending on how accurate and optimal the algorithms are, the designer may end up just following recommendations from A to Z. As the system can’t provide miscellaneous features or the artistic touch required these are areas that mechanical designers can address.
There are many benefits to the machine learning in mechanical engineering including time and costs saved on the operations as well as human labour. As long as the system is constantly checked to ensure the algorithms still address the customer’s requirements the power and influence of the system can only grow. There may be occasions, as with the likes of Google, where the algorithms need tweaked or adjusted to optimize results but this is fairly straight forward.
Autodesk Design Graph
There is strongly funded research ongoing into the likes of Autodesk Design Graph. This system uses algorithms to extract extensive data from 3D designs to classify and categorises components, together with their interactions, to create a living catalog. The system is able to optimise its picking and categorising process and compile the mass of files entrusted to it. For now, if you search through Design Graph for a particular component, such as a joint or a bolted assembly, it is able to extract the searches and return the different options. While people may assume that Design Graph uses tags and titles to identify parts this is not the case. It relies on identifying the shape and the structure to recognise the part.
Machine learning is the future
Many designers are skeptical if not outraged by the possible inclusion of machine learning in design departments. After all, the long term goal of machine learning systems is to override the processes that can be assimilated into an algorithm, reducing the number of jobs and tasks for designers to do. However, on the other hand, the pioneers of machine learning strongly believe that the systems will remove repetitive or boring tasks from a designers brief, allowing them to focus on actual challenges – improving their designs and techniques along the way.